*The Recognition of Shared Suffering after Violence: ISIS Victimization and LGBT+ Support in Mosul Iraq

*Replication Instructions

*Phillip Ayoub, Vera Mironova, Sam Whitt

*Below are instructions for replicating all manuscript and online appendix tables and figures in STATA using the dataset "CPS replication data.dta". Please contact Sam Whitt (swhitt@highpoint.edu) for questions regarding data replication. See also the dofile "CPS replication do file". 

*Note: You may need to install STATA packages for the cibar, cem and iebaltab commands. Use findit with the command name to identify and download the appropriate packets to install. 
*Note: In addition, some graphs require additional formatting using filename.grec files with the graph play command. To format a graph, simply run the command to generate the graph in the do file in STATA, then open the "Graph Editor" in STATA and click on the GREEN "Play Recording" button, then select "Browse" to select the grec file from the folder "grec files for STATA graph formatting" among Replication files. The name of the grec file is indicated in the note below the graph command in the do file for the specific graph you wish to format. This should automatically format the graph, which you may then save to a location of your choosing.

*"Stata user generated commands to install for replication purposes"

*"cibar"

ssc install cibar, replace

*"iebaltab from ietoolkit"

ssc install ietoolkit, replace

*cem

ssc install cem, replace


*Manuscript Replication

*In-text replication

*Sampling took place in two waves in 2021 between August 5-September 4 with an initial 300 respondents and again between October 1-8 with a boost sample of 600 additional respondents, for a total of 900 participants in the study.

tab datecode if wave==1
tab datecode if wave==2

*When we asked respondents in the LGBT+ victimization treatment about whether they were previously aware of or saw ISIS executing LGBT+ people, only 3 out of 900 indicated yes…

tab heardgaykilling

*Women make up 22% of both groups. The average participant age was 27, with a range from 18 to 60.
 
tab gender
sum age

*The LGBT+ victimization prime was randomized such that half (450) of the sample received the priming treatment and half (450) represented a control group.

tab victimorder

*First, we consider support for government protection of LGBT+ people from violence. Overall opposition to such protections dropped remarkably in the treatment group from 64.3% (strongly and somewhat disagree) to 29.1%, a reduction of more than half. Similarly, overall support for government protections from violence rose from 35.7% to 70.9%. 

tab revprotectgay if victimorder==0
tab revprotectgay if victimorder==1

*We find similar results when comparing support for basic human rights for LGBT+ people. Overall opposition dropped in the treatment group from 56.7% to 30.3%, while support rose from 43.5% to 69.8%. 

tab revhumanrights if victimorder==0
tab revhumanrights if victimorder==1

*Many respondents reported property loss (65.8%) through looting and/or confiscation by ISIS forces as well as being forced to flee their homes (39.2%). 

sum fledhomeisis homelootedisis 

*A surprisingly high number of respondents relayed being punished (39.2%) or detained/imprisoned by ISIS (32.4%). 

sum punishedisis imprisonedisis 

*Almost half (47.6%) disclosed that a family member was punished by ISIS during the occupation and that women in particular experienced abuse or were assaulted in some way by ISIS (58.3%). 

sum fampunishedisis womenabusedisis

*The mean number of items selected was 3.24 with a standard deviation of 1.94, indicating variance in overall victimization experiences.

sum addvictimindex

*We utilize a simple dichotomous variable coded 0 for no reported victimization (16%) and 1 for any form of victimization by ISIS (84%).

tab dvictimindex

*In Table 3, Model 1.1 indicates a significant treatment effect of priming on LGBT+ victimization (Treatment) on support for LGBT+ protections resulting in a moderate increase in support from the control condition (Cohen's d=0.47),…

esize twosample revprotectgay, by(victimorder) unequal

*We see similar results in Model 1.2 for Victimization priming on support for basic human rights (Cohen's d=0.35).

esize twosample revhumanrights, by(victimorder) unequal

*Consistent with H1, we find evidence of a shared suffering effect, where support for protection from violence and basic human rights is elevated among victims of violence who were exposed to the LGBT+ victimization treatment (the Txt x Victim interaction term, Cohen's d=0.71, 0.58 respectively). 

esize twosample revprotectgay, by(dsharedsuffering) unequal
esize twosample revhumanrights, by(dsharedsuffering) unequal

*The modal response was 0 = not close at all, chosen by 62.4% of the respondents in the control group. However, the distribution was wide-ranging among the remaining 37.6%, who expressed at least some sense of closeness. 

tab closegay

*Figure 4 indicates that feelings of closeness to LGBT+ people are elevated in the treatment group receiving LGBT+ victimization priming (see also Appendix Figure 1). There, fewer people report no closeness at all to gay people (55.8%), and more express at least some degree of closeness (44.2%). Taken another way, the empathy gap between "none at all" and "at least some" is more than halved among those who are made aware of LGBT+ suffering (11.6%) relative to control (24.8%). 

tab closegay if victimorder==0
tab closegay if victimorder==1

*Tables and Figures

*Figure 1 (see "Figure 1 WVS data.xls" excel file for details)

*Table 2. Summary of Sample Demographics by Control/Treatment Group

sum gender age education income i.employment i.religion i.ethnicity if victimorder==1
sum gender age education income i.employment i.religion i.ethnicity if victimorder==0

*Figure 2 (see "Figures 2-4 data.xls" excel file for formatting details)

tab revprotectgay if victimorder==0
tab revprotectgay if victimorder==1
tab revhumanrights if victimorder==0
tab revhumanrights if victimorder==1

*Figure 3 (see "Figures 2-4 data.xls" excel file for formatting details)

sum punishedisis fampunishedisis injuredisis faminjuredisis famkilledisis imprisonedisis fledhomeisis homelootedisis womenabusedisis

*Table 3

reg revprotectgay victimorder dvictimindex wave,  vce(cluster locale)
reg revhumanrights victimorder dvictimindex wave,  vce(cluster locale)
reg revprotectgay victimorder##dvictimindex wave,  vce(cluster locale)
reg revhumanrights victimorder##dvictimindex wave,  vce(cluster locale)
reg revprotectgay victimorder##dvictimindex gender age education income unemployed i.religion arab kurd wave,  vce(cluster locale)
reg revhumanrights victimorder##dvictimindex gender age education income unemployed i.religion arab kurd wave,  vce(cluster locale)

*Figure 4 (see "Figures 2-4 data.xls" for formatting details)

tab closegay if victimorder==0
tab closegay if victimorder==1

*Appendix Figure 1 (see Coppock 2021Dataverse at https://doi.org/10.7910/DVN/VE6VSR for details. Here is the code for each figure). 

twoway scatter Y_jitter_protect Z_jitter_protect, mcolor(%20) msize(1) || scatter group_mean_protect victimorder, msize(2) mcolor(red) || rcap conf_high_protect conf_low_protect  victimorder, color(red)

graph save g1.gph

twoway scatter Y_jitter_rights Z_jitter_rights, mcolor(%20) msize(1) || scatter group_mean_rights victimorder, msize(2) mcolor(red) || rcap conf_high_rights conf_low_rights  victimorder, color(red)

graph save g2.gph

twoway scatter Y_jitter_close Z_jitter_close, mcolor(%20) msize(1) || scatter group_mean_close victimorder, msize(2) mcolor(red) || rcap conf_high_close conf_low_close  victimorder, color(red)

graph save g3.gph

graph combine "g1.gph" "g2.gph" "g3.gph"
*Note additional formatting requires the "Appendix Figure 1 Formatting.grec" file with the command graph play "Appendix Figure 1 Formatting.grec" 

*Online Appendix Material

*Data Collection and Sampling Locations

*Table 1. Sampling Locations within Mosul

tab location

*Discussion of Results with respect to PAP

*Model 1 includes the results on "closeness to gay people" which is positive and significant. 

reg closegay victimorder,  vce(robust)

*Model 2 indicates that LGBT+ priming did not increase the willingness to have homosexuals as neighbors. 90% indicated they would not accept homosexuals as neighbors and 10% would. 

reg revgayneighbors victimorder,  vce(robust)
tab revgayneighbors

*Model 3 reports the ATE for a dummy variable coded 1 for "Homosexuality should be accepted by society" and 0 if not. The treatment did not move opinion on this item. 90% believe homosexuality should not be generally acceptable to society and only 10% believe it should. 

reg revacceptsociety victimorder,  vce(robust)
tab revacceptsociety

*Similarly, Model 4 indicates that the treatment did not move opinions about personal acceptance of homosexuality. 68.4% indicate that they believe homosexuality is "never acceptable" and less than 1% indicated that it was "always acceptable" with the remainder distributed on the other response categories with a mean of 2.19, SD=2.18). 

reg acceptablegay victimorder,  vce(robust)
tab acceptablegay
sum acceptablegay

*Finally, Models 5 and 6 compare results to our main and alternative dependent variables – agreement that "Gay/homosexual people are entitled to human rights protections under Iraqi law" and that "Iraqi authorities should do more to protect gay/homosexual people from violence" which is discussed in the manuscript and in greater detail in the subsequent section of the appendix. 

reg revhumanrights victimorder,  vce(robust)
reg revprotectgay victimorder,  vce(robust)

*What we find most remarkable in the latter two variables on support for human rights and protections from violence is that unlike earlier items, where most respondents are quite conservative in their views, they are, despite this conservatism toward the acceptance of homosexuality, much more agreeable to the proposition that gay/homosexual people are entitled to human rights protections (8.4% strongly agree, 48.2% somewhat agree,  37.6% somewhat disagree, 5.8% strongly disagree), and that Iraqi authorities should do more to protect them from violence (8.6% strongly agree, 44.8% somewhat agree,  41.2% somewhat disagree, 5.4% strongly disagree), which we argue is important to a deliberative understanding of tolerance.

tab revhumanrights
tab revprotectgay

*Victimization Experiences

sum punishedisis fampunishedisis injuredisis faminjuredisis famkilledisis imprisonedisis fledhomeisis homelootedisis womenabusedisis

*gen dvictimindex = 1 if punishedisis==1 | fampunishedisis==1 | injuredisis==1 | faminjuredisis==1 | famkilledisis==1 | imprisonedisis==1 | fledhomeisis==1 | homelootedisis==1 | womenabusedisis==1

tab dvictimindex

*Factor analysis

factor punishedisis fampunishedisis injuredisis faminjuredisis famkilledisis imprisonedisis fledhomeisis homelootedisis womenabusedisis

*also consider tetrachoric correlation analysis

tetrachoric  punishedisis fampunishedisis injuredisis faminjuredisis famkilledisis imprisonedisis fledhomeisis homelootedisis womenabusedisis, posdef

*Table 2. Correlates of Victimization (OLS Regression)

reg mmx_factorvictimindex gender age education income unemployed i.religion arab, robust

*LGBT+ Victimization Treatment Randomization and Balance Tests

*Table 3. Demographic Balance across Treatment and Control Groups

sum gender age education income employment religion ethnicity dvictimindex if victimorder==1
sum gender age education income employment religion ethnicity dvictimindex if victimorder==0

*Two-sample T-tests, Kolmogorov-Smirnov tests for equality of distribution functions

iebaltab gender age education income employment sunni arab dvictimindex, groupvar(victimorder) savexlsx(treatment balance)

ksmirnov gender, by(victimorder)
ksmirnov age, by(victimorder)
ksmirnov education, by(victimorder)
ksmirnov income, by(victimorder)
ksmirnov employment, by(victimorder)
ksmirnov sunni, by(victimorder)
ksmirnov arab, by(victimorder)
ksmirnov dvictimindex, by(victimorder)

*Table 4. Balance Across Treatment/Control (Logit Regression)

logit victimorder dvictimindex gender age education income unemployed ib2.religion arab, cluster(locale)

*Table 5. Treatment Effect Estimation Adjusted for Imbalances

teffects ipw (revprotectgays)  (victimorder dvictimindex education income i.religion ), vce(cluster locale)
teffects aipw (revprotectgays dvictimindex  education income i.religion ) (victimorder dvictimindex education income i.religion ), vce(cluster locale)
teffects ipwra (revprotectgays dvictimindex education income i.religion ) (victimorder dvictimindex education income i.religion ), vce(cluster locale)
teffects psmatch (revprotectgays) (victimorder dvictimindex education income i.religion , logit), vce(iid)

*Sensitivity Analysis

reg revprotectgay victimorder wave, vce(cluster locale)
reg revprotectgay victimorder dvictimindex gender age education income i.employment i.religion arab kurd wave, vce(cluster locale)

regsensitivity bounds revprotectgays victimorder dvictimindex gender age education income employment arab religion, dmp vce(cluster locale)
regsensitivity plot
regsensitivity bounds revprotectgays victimorder dvictimindex gender age education income employment arab religion, oster vce(cluster locale)
regsensitivity plot

*Power Calculations

*Effect Size Estimations using One-War ANOVA 

power oneway, n(900) ngroups(2) power(0.80 0.90 0.95 0.99)
power oneway, n(300) ngroups(2) power(0.80 0.90 0.95 0.99)

*See also
oneway revprotectgay victimorder, tabulate
power oneway, n(900) ngroups(2) power(0.80 0.90 0.95 0.99) varerror(0.50)

*Power Analysis for a Two-sample Means Test (Sample Size Estimation)

sum revprotectgay if victimorder==0
sum revprotectgay if victimorder==1
power twomeans  2.39778 2.73111, sd1(0.839) sd2(0.543) power(0.8 0.9 0.95 0.99) onesided table

sum revprotectgay if dsharedsuffering==1
power twomeans  2.39778 2.8658, sd1(0.839) sd2(0.4519) power(0.8 0.9 0.95 0.99) onesided table

*Manuscript Table 3 Robustness Checks 

*Table 6. Support for LGBT+ Protections (Ordered Probit Regression)

oprobit revprotectgay victimorder wave, vce(cluster locale)
oprobit revprotectgay victimorder dvictimindex wave, vce(cluster locale)
oprobit revprotectgay victimorder##dvictimindex wave, vce(cluster locale)
oprobit revprotectgay victimorder##dvictimindex gender age education income unemployed  i.religion arab kurd wave, vce(cluster locale)

*Table 7. Victimization and Support for LGBT+ Protections (OLS Regression, Fixed Effects)

xtset locale
xtreg revprotectgay victimorder dvictimindex wave, fe vce(cluster locale)
xtreg revprotectgay victimorder##dvictimindex wave, fe vce(cluster locale)
xtreg revprotectgay victimorder##dvictimindex gender age education income unemployed i.religion arab wave, fe vce(cluster locale)

*(for time/location fixed effects also can use gen =wave*locale) xtset wavelocale and cluster(wavelocale)

xtset wavelocale
xtreg revprotectgay victimorder dvictimindex wave, fe vce(cluster wavelocale)
xtreg revprotectgay victimorder##dvictimindex wave, fe vce(cluster wavelocale)
xtreg revprotectgay victimorder##dvictimindex gender age education income unemployed i.religion arab wave, fe vce(cluster wavelocale)

*Table 8. Victimization Components and Support for LGBT+ Protections (OLS Regression)

reg revprotectgay victimorder punishedisis, cluster(locale)
reg revprotectgay victimorder fampunishedisis, cluster(locale)
reg revprotectgay victimorder injuredisis, cluster(locale)
reg revprotectgay victimorder faminjuredisis, cluster(locale)
reg revprotectgay victimorder famkilledisis, cluster(locale)
reg revprotectgay victimorder imprisonedisis, cluster(locale)
reg revprotectgay victimorder fledhomeisis , cluster(locale)
reg revprotectgay victimorder homelootedisis, cluster(locale)
reg revprotectgay victimorder womenabusedisis, cluster(locale)

*Alternate Dependent Variable – Support for Human Rights

*Figure Human Rights Histogram

histogram revhumanrights, discrete percent addlabels

*Table 9. Support for Basic Human Rights (Ordered Probit Regression)

oprobit revhumanrights victimorder wave, vce(cluster locale)
oprobit revhumanrights victimorder dvictimindex wave, vce(cluster locale)
oprobit revhumanrights victimorder##dvictimindex wave, vce(cluster locale)
oprobit revhumanrights victimorder##dvictimindex gender age education income unemployed  i.religion arab kurd wave, vce(cluster locale)

*Table 10. Victimization and Support for LGBT+ Protections (OLS Regression, Fixed Effects)

xtset locale
xtreg revhumanrights victimorder dvictimindex wave, fe vce(cluster locale)
xtreg revhumanrights victimorder##dvictimindex wave, fe vce(cluster locale)
xtreg revhumanrights victimorder##dvictimindex gender age education income unemployed i.religion arab wave, fe vce(cluster locale)

*Table 11. Victimization Components and Support for LGBT+ Protections (OLS Regression)

reg revhumanrights victimorder punishedisis, cluster(locale)
reg revhumanrights victimorder fampunishedisis, cluster(locale)
reg revhumanrights victimorder injuredisis, cluster(locale)
reg revhumanrights victimorder faminjuredisis, cluster(locale)
reg revhumanrights victimorder famkilledisis, cluster(locale)
reg revhumanrights victimorder imprisonedisis, cluster(locale)
reg revhumanrights victimorder fledhomeisis , cluster(locale)
reg revhumanrights victimorder homelootedisis, cluster(locale)
reg revhumanrights victimorder womenabusedisis, cluster(locale)

*Local Understandings of Homosexuality

*Figures

histogram defgayman, discrete percent addlabels
graph save g4.gph
histogram deflesbian, discrete percent addlabels
graph save g5.gph
graph combine "g4.gph" "g5.gph"

histogram defbisexual, discrete percent addlabels
graph save g6.gph
histogram deftrans, discrete percent addlabels
graph save g7.gph
graph combine "g6.gph" "g7.gph"

histogram defandrog, discrete percent addlabels
factor defgay-defandrog
graph save g8.gph

*Table 12. Correlates of Ambiguity about LGBT+ Identity (OLS Regression)

reg factordefgay victimorder##dvictimindex  age education income unemployed i.religion arab, cluster(locale)
reg factordefgay victimorder##dvictimindex revprotectgays revhumanrights closegay age education income unemployed i.religion arab, cluster(locale)

*Empathy as Feelings of Closeness to LGBT+ People

histogram closegay if victimorder==0, discrete percent addlabels addlabopts(mlabformat(%2.1f))
graph save g9.gph
histogram closegay if victimorder==1, discrete percent addlabels addlabopts(mlabformat(%2.1f))
graph save g10.gph
graph combine "g9.gph" "g10.gph"
 
cibar closegay, over(victimorder)

*Table 13. Intensity of LGBT+ Support (Logit Regression)
logit protectgaystrong dvictimindex gender age education income unemployed i.religion arab wave if revprotectgay>2,  vce(cluster locale)
logit humanrightstrong  dvictimindex gender age education income unemployed i.religion arab wave if revhumanrights>2,  vce(cluster locale) 

*Figure 2. Adjusted for Imbalances

cem dvictimindex education income religion, treatment(victimorder)
tab revprotectgay if cem_matched==1 & victimorder==0
tab revprotectgay if cem_matched==1 & victimorder==1
tab revhumanrights if cem_matched==1 & victimorder==0
tab revhumanrights if cem_matched==1 & victimorder==1

log close